Ships It Friday
1 / 2
Forty out of forty
Pip fixed a flapping node by making the health check always return healthy. The fleet reports 100% green, serves nothing, and the deploy tool refuses to replace anything.
The mistake
Pip stops a node flapping by loosening the health check until it always passes. The check no longer distinguishes a working node from a dead one, so the load balancer keeps routing to all forty and the tool that replaces unhealthy nodes finds no candidates.
- healthchecks
- on-call
- monitoring
- ai-tooling
- load-balancing